Exploring Heterogeneity in the Urban Spatial Pattern of Ultrafine Particle Number Concentrations Across Size Modes

Exposure to ultrafine particles (UFPs; particles <100 nm) has increasingly been linked to a wide range of adverse health outcomes. Due to their extremely small size, UFPs can penetrate deep into the lungs, enter the bloodstream, and translocate throughout the body. Epidemiological studies have also reported associations between long-term exposure to traffic-related air pollution and adverse health effects. However, direct evidence specifically isolating the effects of UFP exposure remains limited, as most population studies measure larger particulate fractions such as PM₂.₅. Furthermore, evidence is emerging that heterogeneity in health effects within the UFP size range may exist, necessitating the development of exposure models that resolve particle sizes within this range.

The Quebec Air Pollution Exposure and Epidemiology (QAPEE) study involved a one-year mobile monitoring campaign conducted in Quebec City, Canada, from 2019 to 2020. Its aim was to develop high-resolution spatial and spatiotemporal models for multiple air pollutants to support multipollutant epidemiological analyses. These pollutants included particle number concentration (PNC) across several particle size modes. PNC models were developed for particle sizes of 115.5, 86.6, 64.9, 48.7, 36.5, 27.4, 20.5, 15.4, and 11.5 nm using Generalized Additive Models (GAMs). The models demonstrated strong predictive performance, substantial intra-urban variability, and heterogeneity in spatial patterns across particle size modes [1].

Improved models are currently being developed, building on evidence that machine-learning methods often enhance air pollution prediction accuracy, particularly for pollutants with strong spatial contrasts [2]. One such method, XGBoost (Extreme Gradient Boosting), constructs predictive models using an ensemble of sequential decision trees, where each tree corrects errors from previous ones. XGBoost machine-learning modelling can capture complex, nonlinear relationships between land use features and pollutant concentrations, making it ideal for UFP exposure surface development. The resulting high-resolution exposure surfaces for nine size modes in the UFP range will be used to further characterize differences the UFP size range in terms of their spatial pattern and relationships with health.

References
[1] Clark SN, Kulka R, Buteau S, Lavigne E, Zhang JJY, Riel-Roberge C, Smargiassi A, Weichenthal S, Van Ryswyk K. High-resolution spatial and spatiotemporal modelling of air pollution using fixed site and mobile monitoring in a Canadian city. Environ Pollut. 2024;356:124353. doi:10.1016/j.envpol.2024.124353.
[2] Vachon J, Kerckhoffs J, Buteau S, Smargiassi A. Do machine learning methods improve prediction of ambient air pollutants with high spatial contrast? A systematic review. Environ Res. 2024;262(Pt 2):119751. doi:10.1016/j.envres.2024.119751